A method for predicting concentration of dissolved gas in transformer oil based on improved adaptive filtering algorithm

Abstract

The application discloses a method for predicting the concentration of dissolved gas in transformer oil based on an improved adaptive filtering algorithm, and the steps of the method comprise the following steps: step 1, obtaining a historical sequence of the concentration of dissolved gas in oil; step 2, moving a sliding window and averaging to construct a new sequence; step 3, on the basis of the newly constructed sequence, obtaining the weight variation trend of previous iterations by using the improved adaptive filtering algorithm, averaging each weight component in pairs according to the order without repetition and without interval, so as to construct a sequence of weight components, fitting the sequence of weight components by using an exponential-based mathematical model, and obtaining the final weight component by taking the limit of the obtained expression; and step 4, predicting the future gas concentration according to the weight vector. Through the prediction of the concentration of dissolved gas in oil, the application can better realize the fault early warning of the transformer and help the safe and reliable operation of the power system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power equipment state evaluation, and particularly relates to a transformer oil dissolved gas concentration prediction method based on an improved adaptive filtering algorithm. BACKGROUND

[0002] The transformer is an extremely important device in the power system, and its safe and reliable operation is related to the stability of the entire power system, so the significance of studying it is self-evident.

[0003] The transformer is mainly divided into oil-immersed transformers and dry transformers. For the former, the gas content dissolved in the transformer oil can be monitored online, so that the fault type of the transformer can be reflected through the nonlinear mapping relationship between the transformer fault phenomenon and the fault type, and the fault diagnosis of the transformer can be realized. Although this method has greatly improved the fault diagnosis efficiency and health management level of the transformer, it lacks a certain predictability. If the gas concentration can be predicted on the basis of fault diagnosis, it will greatly benefit the fault warning of the transformer and prevent the occurrence of risk accidents.

[0004] For gas prediction, there are many prediction algorithms, such as LSTM, SVM and gray theory, etc. However, these algorithms have the shortcomings of complex mechanism or poor applicability, and adaptive filtering is not only simple but also has good adaptability, so it is used for the prediction of dissolved gas in transformer oil.

[0005] For the past practice of dissolved gas prediction in oil, each component sequence is individually regressed, and the coupling effect between each component is ignored. The present patent fully considers this influence. However, the traditional adaptive filtering can only adapt to a single sequence, and the calculation amount greatly increases when there are many weights. The present patent ingeniously designs a "multi-level" and "limited iteration" adaptive filtering, i.e. an improved adaptive filtering, to obtain the weight components in the iteration process. Since the iteration will eventually converge, the processed weight components are taken to the limit in the form of an exponential, so as to obtain the final weight, and then the gas concentration is predicted. SUMMARY

[0006] Based on the above technical problems, the purpose of the present application is to overcome the shortcomings of the traditional method, such as complex mechanism, poor adaptability and the need for a large number of training samples, so as to provide a transformer oil dissolved gas concentration prediction method based on an improved adaptive filtering algorithm.

[0007] To solve the above calculation problem, the present application adopts the following technical scheme:

[0008] A transformer oil dissolved gas concentration prediction method based on an improved adaptive filtering algorithm, comprising the following steps:

[0009] Step 1, obtaining a historical sequence of dissolved gas concentration in oil;

[0010] Step 2, moving a sliding window and averaging to construct a new sequence;

[0011] Step 3, on the basis of the newly constructed sequence, using an improved adaptive filtering algorithm to obtain the weight change trend of the previous several iterations, averaging each weight component in order without repetition and interval, thereby constructing a sequence of weight components, using an exponential-based mathematical model to fit the weight component sequence, and obtaining the final weight component by taking the limit of the obtained expression;

[0012] Step 4, predicting the future gas concentration according to the weight vector.

[0013] Further, the dissolved gas in the oil in step 1 refers to H2, CH4, C2H2, C2H4 and C2H6; wherein the sampling interval of the historical sequence of dissolved gas concentration in oil is uniform.

[0014] Further, the step 2 includes:

[0015] Step 2.1, the size of the sliding window is adjusted according to the length of the sequence;

[0016] Step 2.2, averaging each sliding window to construct a new sequence.

[0017] Further, the step 3 includes the following sub-steps:

[0018] Step 3.1, based on the improved adaptive filtering algorithm, constructing a weight component and gas concentration prediction model,

[0019] For the improved adaptive filtering, the model construction is as follows:

[0020]

[0021] wherein m is the total number of gas species, i is the gas species, n is the number of items of each gas concentration, k is the index of the historical time concentration corresponding coefficient of any kind of gas, N i，j-k is the actual value of the i-th gas at the j-k time, is the predicted value of the i-th gas at the j time, w i，k is the weight of the i-th gas concentration at the j-k time;

[0022] Step 3.2, setting the iteration termination condition, i.e. setting the iteration number α to reach the upper limit α limit As the iteration termination criterion, initialize α = 0;

[0023] Step 3.3, set learning rate lr and initialize all weights wherein w i,k (α) is the weight of the concentration of the ith gas at the j-k moment in the a iteration;

[0024] Step 3.4, initialize iteration variable j, wherein n < j ≤ p, p is the latest moment of known concentration;

[0025] Step 3.5, according to calculate wherein, is the predicted value of the ith gas at the j moment;

[0026] Step 3.6, record the error between the true value N i,j and the predicted value , that is,

[0027] Step 3.7, adjust the weight w i,k according to the error (α+1) = w i,k (α) + 2·lr·error·N i,j-k , and record the weight w i,k at this time (α+1) , and let α = α + 1;

[0028] Step 3.8, judge whether the iteration number exceeds the set value, if it exceeds, turn to step 3.9; otherwise, continue iteration until the iteration number exceeds the set value to obtain the sequence of the weight component;

[0029] Step 3.9, for each sequence of weight components, take the average of each two in order without repetition and interval, thereby constructing a new sequence of weight components;

[0030] Step 3.10, for the change trend of each weight component, parameter estimation is performed in the form of y = a + be -cx , wherein x represents the average of the iteration numbers of two times, and y represents the average of the weight components of each two iterations;

[0031] Step 3.11, extremum is obtained for the obtained parameter expression, thereby obtaining the final weight component.

[0032] Further, the step 4 comprises:

[0033] Step 4.1, on the basis of the obtained weight component, substitute the weight component and the gas concentration prediction model to predict the future gas concentration.

[0034] Compared with the prior art, the present application has the following beneficial effects:

[0035] (1) The coupling effect between the components of dissolved gas in oil is fully considered, making the prediction more accurate;

[0036] (2) The adaptive filtering was improved and made applicable to the processing of multiple sequences. In addition, the problem of the precision threshold being difficult to determine was also solved.

[0037] (3) In the iterative process, especially when the model contains multiple weights, the problem is particularly time-consuming. By performing exponential parameter estimation on the weight component sequence, the weights can be located quickly, which improves the speed and efficiency of the solution. Attached Figure Description

[0038] Figure 1 This invention provides a schematic flowchart of a method for predicting dissolved gas concentration in transformer oil based on an improved adaptive filtering algorithm. Detailed Implementation

[0039] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, embodiments and comparative examples.

[0040] Example

[0041] like Figure 1 As shown, this embodiment provides a method for predicting the concentration of dissolved gases in transformer oil based on an improved adaptive filtering algorithm, including the following steps:

[0042] Step 1: Obtain the historical sequence of dissolved gas concentrations in the oil;

[0043] Step 2: Move the sliding window and construct a new sequence on average;

[0044] Step 3: Based on the newly constructed sequence, an improved adaptive filtering algorithm is used to obtain the weight change trend of the previous several iterations. The weight components are averaged in pairs without repetition or gap to construct the weight component sequence. An exponential-based mathematical model is used to fit the weight component sequence, and the limit of the obtained expression is obtained to obtain the final weight components.

[0045] Step 4: Based on the weight vector, predict the future gas concentration.

[0046] In the above embodiments, the dissolved gases in the oil in step 1 refer to H2, CH4, C2H2, C2H4, and C2H6, etc.; and the sampling intervals should be kept as uniform as possible.

[0047] In a further preferred embodiment, step 2 includes:

[0048] Step 2.1: Adjust the size of the sliding window according to the length of the sequence;

[0049] Step 2.2, average each sliding window to construct a new sequence.

[0050] In the above embodiment, the size of the sliding window can be adjusted according to the length of the sequence; the difference between the length of the sequence and the size of the sliding window is preferably not greater than 10;

[0051] In a further preferred embodiment, step 3 comprises the following sub-steps:

[0052] Step 3.1, based on the improved adaptive filtering algorithm, construct the weight component and the gas concentration prediction model, for the improved adaptive filtering, the model construction is as follows

[0053]

[0054] Wherein, m is the total number of gas species, i is the gas species, n is the number of items of each gas concentration, k is the index of the historical time concentration corresponding coefficient of any kind of gas, N i，j-k is the actual value of the i-th gas at the j-k time, is the predicted value of the i-th gas at the j time, w i，k is the weight of the i-th gas concentration at the j-k time;

[0055] Step 3.2, set the iteration termination condition, that is, set the iteration number a to reach the upper limit a limit As the iteration termination criterion, initialize a = 0; In this embodiment, unlike the adaptive filtering, the precision is used as the iteration termination criterion, but the iteration upper limit is set, for example, the iteration upper limit is set to 1000 times, and the initial iteration number is set to a = 0;

[0056] Step 3.3, set the learning rate lr and initialize all weights Wherein, w i,k (α) is the weight of the i-th gas concentration at the j-k time in the a-th iteration; given the predicted time j, set the weight n x m, the learning rate is approximately 1, for example, 0.9, and all weights wi ,k = 1 / nm;

[0057] Step 3.4, initialize the iteration variable j, wherein n < j ≤ p, p is the latest time of known concentration;

[0058] Step 3.5, according to Calculate Wherein, is the predicted value of the i-th gas at the j time;

[0059] Step 3.6, record the true value N i,j and the predicted value The error, i.e.

[0060] Step 3.7: Adjust the weights w based on the error. i,k (α+1) =w i,k (α) +2·lr·error·N i,j-k And record the weight w at this time. i,k (α+1) And let α = α + 1;

[0061] Step 3.8: Determine if the number of iterations exceeds the set value. If it does, proceed to step 3.9; otherwise, continue iterating until the number of iterations exceeds the set value to obtain the sequence of the weight component.

[0062] Step 3.9: For each sequence formed by the weight components, calculate the average of each pair of components in order without repetition or gap, thereby constructing a new sequence of weight components.

[0063] Step 3.10: For the changing trend of each weight component, use the form y = a + be -cx Parameter estimation is performed in the form of , where x represents the average number of iterations between two iterations, and y represents the average weight component between every two iterations; its value is 1.5, 2.5, 3.5, ..., where y represents the average weight component between every two iterations, i.e. The superscript indicates the iteration number;

[0064] Step 3.11: Find the extreme value of the obtained parameter expression to obtain the final weight components.

[0065] In the above embodiments, the improved adaptive filtering differs from traditional adaptive filtering. For any gas, its future development state is not only related to the development of the gas itself, but also affected by other gases. Therefore, the historical concentrations of other gases are coupled into the expression of the mathematical model to be established. However, traditional adaptive filtering can only be applied to a single time series and is helpless for multiple time series. The improved adaptive filtering can solve the problem that traditional adaptive filtering cannot handle during the iteration process by synchronously moving all time series.

[0066] Specifically, when using adaptive filtering for prediction, assuming there are n weights, then for the i-th gas, at time j,

[0067]

[0068] Without loss of generality, assume i = 1, j = 10; n = 3, then

[0069]

[0070] For the improved adaptive filter, the model is constructed as follows

[0071]

[0072] where m is the gas type, here taken as 5, and it is assumed that j = 10; n = 3, then

[0073]

[0074] In the method, the basis of the mathematical model uses equation (3);

[0075] Note that the number of weights increases dramatically;

[0076] In a further preferred embodiment, step 4 comprises:

[0077] Step 4.1, on the basis of the obtained weight vector, the future gas concentration is predicted.

[0078] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application should be equivalent replacement methods, and are included in the protection scope of the present application.

Claims

1. A method for transformer oil dissolved gas concentration prediction based on improved adaptive filtering algorithm, characterized in that, The method comprises the following steps: Step 1, obtaining a historical sequence of dissolved gas concentration in oil; Step 2, moving a sliding window and averaging to construct a new sequence; comprising: Step 2.1, the size of the sliding window is adjusted according to the length of the sequence; Step 2.2, averaging each sliding window to construct a new sequence; Step 3, on the basis of the newly constructed sequence, using an improved adaptive filtering algorithm to obtain the weight change trend of the previous several iterations, and averaging each weight component in order without repetition and interval to construct a sequence of weight components, using an exponential-based mathematical model to fit the weight component sequence, and taking the limit of the obtained expression to obtain the final weight component; comprising the following sub-steps: Step 3.1, based on the improved adaptive filtering algorithm, constructing a weight component and gas concentration prediction model, For the improved adaptive filtering, the model construction is as follows: wherein, m is the total number of gas species, i is the number of gas species, n is the number of terms for the concentration of each gas, k is the index of the corresponding coefficient of the concentration at the historical moment for any gas species, N i，j-k is the actual value of the i th gas at the j-k th moment, is the predicted value of the i th gas at the j th moment, w i，k is the weight of the concentration of the i th gas at the j-k th moment. Step 3.2, set the iteration termination condition, that is, set the number of iterations α reaching the upper limit α limit As the iteration termination criterion, initialize α=0 ; Step 3.3, Set learning rate lr and initialize all weights where, is the weight of the concentration of the α th gas at the j-k th time instant at the i th iteration. Step 3.4, initialize iteration variable j wherein, n<j≤p , p is the latest time with known concentration; Step 3.5, according to calculate ,in, For the first i The gas in the first j The predicted value at any given time; Step 3.6, record true value error from predicted value i.e. ; Step 3.

7. Adjust the weights according to the error and record the weights at this time and let α = α + 1; Step 3.8, judging whether the iteration number exceeds the set value, if yes, turning to step 3.9; otherwise, continuing iteration until the iteration number exceeds the set value to obtain the sequence of the weight component; Step 3.9, for each sequence of weight components, averaging in order without repetition and interval to construct a new sequence of weight components; Step 3.

10. For each weight component, the parameter estimates are made in the form of y=a+be -cx where x represents the average of the number of iterations between two times, y and where the average of the weight component between two times is represented by Step 3.11, taking the extreme value of the obtained parameter expression to obtain the final weight component; Step 4, predicting the future gas concentration according to the weight vector.

2. The method for predicting the concentration of dissolved gases in transformer oil based on an improved adaptive filtering algorithm according to claim 1, characterized in that: The dissolved gas in the oil in step 1 refers to H2, CH4, C2H2, C2H4 and C2H6; wherein the sampling interval of the historical sequence of dissolved gas concentration in oil is uniform.

3. The method for predicting the concentration of dissolved gases in transformer oil based on an improved adaptive filtering algorithm according to claim 1, characterized in that: The step 4 comprises: Step 4.1, on the basis of the obtained weight component, substituting the weight component and the gas concentration prediction model to predict the future gas concentration.

Images